Once a project has a concept, a client, a script, and an overall direction, the next challenge is translating those words into something visual. This is the moment when a project begins to take its actual form, when the creative direction moves from a description into something you can see, evaluate, and show to a client. Style frames are the tool for this. They are high-quality images that represent the visual language of a project, created before any animation or editing begins, and used to establish and get approval for the overall look and feel.
- Style frames define the visual direction of a project and are typically presented for client approval before production work begins. They are not final deliverables but high-quality representations of where the creative direction is heading.
- Moving from a written visual description in an AV script to an image generator prompt requires an intermediate step through an LLM, which helps translate production language into the kind of detailed, atmosphere-rich description that image generators respond to most effectively.
- Generated images almost always need editing in a tool like Photoshop before they are ready to present as style frames. AI gets you to a strong starting point quickly, but the polish and logical coherence of the image require additional work.
This lesson is a preview from our Generative AI Certificate Online. Enroll in a course for detailed lessons, live instructor support, and project-based training.
This is a lesson preview only. For the full lesson, purchase the course here.
The workflow described here connects three tools in sequence: an LLM for prompt development, an image generator for initial visual production, and an image editing program for refinement. Understanding how each of these contributes to the final style frame, and where the creative judgment lives in the process, is what makes the approach repeatable and professional rather than just fast.
Starting with the Visual Descriptions in an AV Script
Your AV script contains visual descriptions: notes about what should appear on screen during each audio beat. These are planning notes, not image prompts. They describe production intent but typically lack the visual specificity that image generators need to produce something useful. A note like "atmospheric corridor with a sense of mystery" is a useful planning direction, but will produce generic results in an image generator without further development.
The first step is to take one of these visual descriptions and use it as the basis for a more detailed image generation prompt. Choose a description that represents a key visual moment in the project: a scene that will define the look and feel of the piece for the client, or a moment that captures the emotional core of the work. You do not need to generate style frames for every scene at this stage. You need enough coverage to communicate the creative direction clearly, which usually means two to four strong images that show different aspects of the visual world you are building.
Identify the description you are going to work from and note anything important about the brand context: the client's visual identity, the emotional tone the brand wants to project, and any specific visual constraints or preferences that have come out of the brief or the project development process so far. All of this will go into the prompt development step.
Using an LLM to Develop an Image Prompt
Open your LLM of choice and set up the context for your prompt request. Tell it what you are working on: in this case, you are a motion designer developing style frames for a client project, and you want help writing prompts for an image generator. Give it the brand guide or the relevant elements from the brand guide, the visual description from your AV script, and any additional context about the project's visual direction.
Ask the LLM to generate several brand variations on the visual tone. This is useful because even within a single project brief, there is often more than one legitimate visual direction. Presenting a client with two or three distinct visual approaches gives them something meaningful to choose from, and it gives you a clearer picture of the range of possibilities before you commit to developing one direction in depth. You might ask for three variations: one that stays closer to the brand's existing visual language, one that pushes the aesthetic in a more unexpected direction, and one that finds a hybrid between the two.
Once you have a sense of which direction you want to develop, ask the LLM to write a full image prompt for that direction. Specify the elements you want the prompt to include: the mood, the composition, the camera angle, and the visual style. Ask for the prompt to be written in paragraph form rather than as a bulleted list, since image generators generally respond better to flowing descriptive language than to itemized specifications. And ask the tool to follow up with questions before producing the final version of the prompt.
Answering Follow-up Questions to Sharpen the Prompt
The follow-up question step is worth taking seriously. It is easy to skip in the interest of speed, but it consistently produces better image prompts when you do it. The kinds of questions a well-prompted LLM will ask are exactly the kinds of decisions that affect what the image generator produces: should the key subject fill the frame as a hero object, or sit within a larger environment? Should textures be realistic or stylized? Should the composition leave negative space for future animation? What should the camera feel like: grounded and naturalistic, or slightly unnatural and otherworldly?
These are questions you may not have consciously decided before being asked. Answering them forces you to make specific creative choices that then get built into the prompt. The result is a more directed, more intentional image generation experience. You are not simply asking the tool to make something that looks like the general direction. You are asking it to make something with specific compositional logic, specific textural qualities, and specific atmospheric properties that you have decided in advance.
After answering the questions, let the LLM produce the final image prompt. Read it through before moving on. You are looking for language that captures the atmosphere and visual identity you have in mind, with enough specificity to guide the image generator meaningfully. If it is still too generic, ask the tool to revise it with more detail in the areas that matter most to you.
Choosing an Image Generator and Adjusting the Settings
The image generator you use will significantly affect the look of the output. Different tools have different visual signatures. Midjourney is well known for producing images with a distinctive stylized quality that works particularly well for dark, atmospheric, or conceptually unusual subject matter. Adobe Firefly is tightly integrated with the Adobe ecosystem, produces commercially licensed imagery by default, and offers a wide range of stylistic options. Leonardo AI has multiple model options that suit different visual styles and use cases.
For any given project, the choice of tool should be influenced by the visual direction you are pursuing. If you are working toward something moody, surreal, or highly stylized, Midjourney often produces images that feel right in a way that other tools do not. If you are working within an Adobe production pipeline and need imagery that is commercially clean, Firefly is a natural choice. In practice, it is worth generating results in multiple tools and comparing the results before settling on a direction.
Once you are in your tool of choice, spend time on the settings before generating. Most image generation tools offer controls for visual style, stylization level, variety in the outputs, aspect ratio, and model version. Setting the aspect ratio to match your intended output format before generating saves time and produces images that are actually usable in your production context, rather than needing to be cropped or extended. For motion graphics and video work, landscape formats at standard video ratios are usually the right starting point.
Running the Prompt and Evaluating the Results
With the prompt loaded and the settings configured, generate your first set of images. Most tools produce three or four options per generation. Your job at this point is not to find a perfect image but to evaluate the direction and decide what to do next. Look at each output and ask whether it is moving toward what you have in mind, whether the visual language feels right for the project, and whether any of the specific images are interesting enough to develop further.
It is common for the first generation to be promising but not quite there. Maybe the mood is right, but the composition is not. Maybe the stylization level is close, but pushed a little too far in one direction. Maybe one of the four outputs is interesting in a way you did not anticipate and opens up a direction worth exploring further. All of these are normal outcomes, and all of them point toward what to do next.
If the outputs are going in an interesting direction, try running the same prompt again with adjusted settings. In Midjourney, for example, changing the stylization level shifts how interpretive the tool is with your prompt, and changing the variety setting affects how different each of the four outputs will be from one another. In Firefly, adjusting the visual intensity or the lighting option can shift the mood considerably. These adjustments are fast to make and often produce meaningfully different results from the same underlying prompt.
Developing Multiple Visual Directions for Comparison
If you generated multiple brand direction variations from the LLM step, this is the point where you can develop image prompts for more than one of them. Producing style frame references for two or three distinct visual directions gives you something valuable: a set of clearly differentiated options to evaluate and to show to a client.
In a professional presentation context, showing a client three clearly different visual directions, each with two or three supporting images, allows them to make a real choice rather than simply approving or rejecting a single option. It also helps clarify what elements of the brief were driving each direction and what the visual logic of each approach is, which is useful information for developing the chosen direction further after the presentation.
- Generate at least two to three images for each direction you want to present, so the client sees enough to understand the visual language rather than reacting to a single image.
- Label each direction clearly and note the key visual qualities that define it, so the presentation is not just a collection of images but a legible set of options with distinct identities.
- Do not present images that you are not able to defend. If an image has elements that are confusing, structurally wrong, or disconnected from the project intent, fix those before presenting or generate a better alternative.
Fixing and Refining Generated Images Before Presentation
Generated images almost never arrive ready to present without some editing. The most common issues are structural inconsistencies: elements that are in the wrong place, details that do not make physical or spatial sense, proportions that are off, or visual elements that were part of the prompt but appeared in an unexpected or illogical way. These are not failures of the tool. They are a normal part of the generative AI output process, and they are one of the reasons the workflow includes an editing step.
Bringing the generated images into Photoshop or another image editing program lets you fix what does not work while keeping the overall visual language intact. You might correct a compositional element that is structurally wrong. You might adjust the overall color tone to better match the brand direction. You might crop the image differently to emphasize a more interesting part of the composition. You might use painting or retouching tools to clean up areas where the generation was not consistent.
The goal of the editing step is not to radically alter the image. If the image needs to be fundamentally rebuilt to work, it is probably better to go back and generate a new one with a more refined prompt. The editing step is for polish and coherence: making the image work as a professional, presentation-ready piece rather than as an interesting but rough generated output.
What This Process Gives You and What It Does Not
This workflow, moving from an AV script through LLM-assisted prompt development to image generation and editing, gives you a range of high-quality visual references for style frames significantly faster than building those frames from scratch in a design program. In a creative industry context where the pace of client presentations and the pressure to show options quickly is real, that speed advantage is meaningful.
What this process does not give you is a substitute for creative judgment. The LLM can help you develop a prompt, but it does not know which of the three brand directions you explored is actually the right one for this client and this project. The image generator can produce striking imagery, but it does not know whether those images genuinely serve the project's communication goal. The editing step can fix structural issues, but it cannot resolve a fundamental mismatch between the generated imagery and what the project actually needs.
Those decisions are yours, and they require the same design thinking, understanding of the brief, and knowledge of the client that any creative work requires. What the AI tools in this workflow do is take the mechanical work out of getting to a starting point, so that the creative energy you bring to the project is concentrated on the decisions that actually matter.
